# Prophetero Loop Lab GLOBAL 0.3
Find repeatable cuts in supported local videos, compare transition styles and inspect the result before exporting. Works with LTX, H3, camera footage and other sources; it is not tied to a generation model.
Start with Recommended: a clean picture cut and independent short sound smoothing. Suggested cuts include first/last frame previews. Fine-tune either endpoint directly on its card with native-frame -1/+1 buttons, or reset to the original suggestion. Manual edits invalidate the earlier score; the other suggestions remain untouched.
Compare transitions opens three configurable preview cards from the same cut and sound settings. Try direct cut, crossfade, black/white dips, left/right wipes, opening iris or blur. Render all three sequentially or one version, inspect three cycles, then explicitly apply your choice. An effect is a visible scene transition, not a guarantee of natural continuous motion.
Picture and sound stay independent. Sound offers follow-video, smoothing, direct cut or mute. Full exports retain recorded source-generation tags where supported and save an adjacent .loop.json processing recipe. Originals remain unchanged; output picture is re-encoded as H.264, not losslessly copied.
Includes a custom node/workflow, installer and eight helper languages (EN default). Local CPU/FFmpeg processing; no AI generation model required. Install into the ComfyUI backend actually used by your application, restart it fully, and load the updated workflow. A JSON import alone does not install the node or dependencies.
Credits: Prophetero provides the workflow integration, controls, installer and review interface. Media processing and effects use [FFmpeg](https://ffmpeg.org/) through [imageio-ffmpeg](https://github.com/imageio/imageio-ffmpeg); stream inspection uses [PyAV](https://github.com/PyAV-Org/PyAV); local web integration uses [aiohttp](https://github.com/aio-libs/aiohttp), hosted by [ComfyUI](https://github.com/Comfy-Org/ComfyUI). Those projects retain their own terms and are not claimed as Prophetero-authored engines. No FFmpeg binary, AI model or Python dependency is bundled. See THIRD_PARTY.md and LICENSE.txt.
Test edition: heuristic point ranking and automated checks do not replace inspecting your own clip, motion and sound before publishing.
Description
# 0.3.0 — 2026-11-03 — Transitions and candidate fine tuning
- Six additional picture methods: black/white dips, left/right wipes, opening iris and blur.
- Three configurable transition-preview cards using the same source cut and sound settings; explicit sequential rendering and explicit application.
- Per-candidate native-frame start/end stepping, refreshed endpoint pictures, original-point reset and invalidated scores after manual edits.
- Coalesced frame updates, stale-response suppression, no automatic full-video rendering during stepping.
- Independent sound modes, legacy widget positions/defaults, source metadata and processing recipes retained.
- English-default GLOBAL workflow, eight-language helper/transition names and revised third-party credits.
- Added real-media timing/endpoint/effect/audio tests, frame API validation and UI race/comparison tests.
Effects make visible transitions; they do not guarantee a natural seamless loop. Cut with short sound smoothing remains Recommended. Exports re-encode picture. This is a test edition pending your own ComfyUI/SwarmUI video-and-sound trial.
# 0.2.0 — 2026-10-31
- Independent sound transition controls: legacy follow-video, equal-length seam smoothing, direct cut and mute.
- Recommended preset: direct picture cut with a short audio seam treatment; no picture crossfade ghosts introduced by this preset.
- Colour and signed temporal comparisons, predicted seam step and native-frame refinement replace tiny grayscale/motion-amount-only ranking.
- Correct exclusive end-frame coordinates and per-candidate first/last previews.
- Apply and preview combines candidate selection and three-cycle inspection.
- Existing widget positions and old graph defaults preserved; new sound inputs are optional.
- Eight-language labels and instructions; original generation metadata remains preserved.
- Added exact sample-count/middle-alignment, fractional-rate, independent-picture, mute/silent-source, native-frame ranking and legacy-node tests.
Limitations: direct cuts may still jump; rankings are heuristics. Short reflected audio contexts may colour speech, music or transients. Final H.264 video is re-encoded; review real material before publishing.
# 0.1.0 — 2026-10-31 — Test edition
- Local video drag/drop and file selection in a dedicated ComfyUI node.
- Bounded appearance/motion sampling for three suggested cut pairs; editable points.
- Start/end frame comparison, frame-sized field steps and three-cycle preview.
- Tail-to-head video blend with sample-counted audio fades; direct-cut alternative.
- Single-cycle, repetition-count and target-duration exports at the source size/rate.
- H.264/AAC output, source fingerprint and .loop.json processing recipe.
- Existing recognized generation metadata recorded separately from loop settings.
- English-default workflow and eight-language helper controls/notes; EN/HU guides.
- Backend locator supports direct ComfyUI, portable and SwarmUI layouts.
This is a testing package. Source/output checks and automated media tests do
not replace a full user trial in the running ComfyUI/SwarmUI interface.